The Morning 10

What

Brent trades at 90.84 this morning, down 6.1% against Friday's 96.78 settle, with a session low of 89.86 — roughly ten percent below the 101.16 high printed Thursday. The move came after Iran reportedly signalled it would suspend attacks for as long as the American pause on strikes holds. Friday's US session had already voted the same way in a quieter register: the oil ETF closed at 136.69, down 2.0% on the day and still below both the 142 shelf and the ~151 crisis highs printed in May.

If

Brent holds under $91 into the US session and the front of the curve keeps closing the gap to the ~$80 December contract, the twenty-dollar backwardation deflates from the front — which is normalization, not shortage.

Why

This is the resolution of the watch-line carried since Wednesday: crisis is when the back of the curve believes the front. It never did. Spot ran to $101 while December sat near $80, and this morning spot moved toward December rather than the reverse. Every asset that was sold for a discount-rate reason last week was sold on the wrong half of that curve.

Then

The 50-day at 83.97 and the 200-day at 83.73 sit close together beneath the price — that band, not $100, is now the level that says whether the entire summer risk premium comes out or only this week's.

What

Nasdaq futures are up more than 1.5% pre-market and Asia closed mostly higher. That is a mirror image of Friday, when the Nasdaq-100 proxy fell 1.12% to 684.23 while the Dow proxy rose 0.48% and the S&P proxy closed up 0.10% at 738.93. The index that lost the most to the rate-and-war trade is the one being bought back first.

If

The gap holds through the cash open rather than fading into it — Friday's Intel gap is the cautionary example of what a faded open looks like this month.

Why

If the flush was a discount-rate event and the discount rate is being repriced by oil, the highest-duration index should recover first and hardest. That is what futures are saying. It is also the cheapest possible explanation, and it will be tested twice this week by the Fed and by four capex guides.

Then

684.23 was Friday's close, under both the 691.96 test and the 694 shelf that has now been probed three times since mid-June. Reclaiming 694 turns the third test into a triple bottom; failing from a green open turns it into a break.

What

On the day the semiconductor complex fell four percent, ten of the eleven SPDR sector funds closed higher. Technology was the sole decliner at −1.44%; Materials rose 1.93%, Real Estate 2.22%, Consumer Staples 1.11%, Financials 0.86%. Over five sessions the leaderboard is Industrials +2.50%, Utilities +2.25%, Materials +1.85%, Real Estate +1.43% — against Discretionary −4.78% and Communication Services −4.16%. And all of this happened on the session after new tariffs on sixty trading partners were announced.

If

The green-sector breadth survives a day when Tech is also green — that is rotation broadening rather than rotation hiding a de-risking.

Why

A market unwinding risk does not pay Materials, Real Estate and Staples on the same day it sells semis by four percent, and it does not shrug at a sixty-country tariff announcement. This was capital moving inside the index, not leaving it — the distinction that separates a rotation from a drawdown.

Then

Discretionary and Communication Services carry the week's damage almost alone, and both are single-name stories — Tesla and Alphabet. Watch whether the laggards mean-revert on the oil break or whether the damage stays company-specific.

What

The broad semiconductor fund closed at 527.01, down 4.40%, below the 530–532 band this letter has tracked as the structural line. The small-cap semis fell hardest at −5.55% to 494.31; the fabless-heavy cut fell 3.75%; the large-cap benchmark fell 3.27% to 561.19. Software went the other way on the same session: the software fund rose 1.01% to 87.98 and the cloud fund rose 2.44%. Our own Rubin 100 fell 3.84% on the day yet still holds a 0.35% weekly gain — and sits 16.3% below where it was a month ago.

If

The green open reclaims 530–532 and holds it into the close, the break reads as a one-session overshoot on a war premium that has since deflated.

Why

The band mattered because it was the first level where the build-out trade and the tape disagreed. Losing it on the day of a rate flush, while software rose, says the market is discriminating within technology rather than exiting it — paying for recurring revenue and selling the capital cycle.

Then

A reclaim without follow-through selling is the cleanest possible answer. Failure to reclaim on a day when oil fell six percent and futures rose would be the more serious message — it would mean the semi problem is not the discount rate.

What

Thursday night Intel printed $16.13 billion in revenue against a $14.4 billion consensus, the fastest growth since 2011, with data-centre and AI up 59%, and traded near $112.70 after hours. On Friday it closed at 92.32, down 7.89% on the day — roughly eighteen percent below the after-hours print, and 9.09% lower on the week.

If

The stock stabilises above $92 on a green tape with oil down six percent, the round trip reads as position unwind rather than a verdict on the numbers.

Why

This is the reaction function, and it is the first trigger of the switch checklist scored in Sunday's Weekly. A double beat with the strongest growth line in fifteen years was sold, on the same day the market paid the software billers. What is being repriced is not whether the demand exists — it is who gets to keep the earnings from it.

Then

Three more capex spenders report inside 48 hours starting Wednesday. If beats with raised guides keep being sold, the trigger stops being an Intel story and becomes the regime.

What

CXMT, China's largest DRAM maker, skyrocketed roughly 470% in its Shanghai STAR Market debut this morning after Asia's largest IPO of the year, carrying an $85.5 billion valuation into the open, with proceeds earmarked for mass-producing memory wafers. On Friday, Micron fell 6.99% to 920.95 and the Memory layer of our own Rubin build-out fell 7.62% on the day — down 29.3% over the month, and still the layer our regime model tags green.

If

The debut premium holds for more than a session, Chinese memory capex gets a public-market funding channel it did not have — at precisely the moment Western memory pricing is the most crowded long in the complex.

Why

A 470% first-day move is a scarcity signal about Chinese listings, not a demand signal about DRAM — one strategist on the tape this morning read it as technical and put the memory cycle near a short-term peak. But the capital raised is real, and it funds wafers.

Then

The memory layer is where our build-out model and the tape disagree most sharply: green regime, minus twenty-nine percent in a month. One of the two is wrong, and the resolution is a pricing question, not a chart question.

What

Moonshot AI's full open-weights checkpoint for Kimi K3 is scheduled for release today: 2.8 trillion parameters in a mixture-of-experts design, a one-million-token context window, and reasoning on by default. The model has been live through the app and API since 16 July, and the first independent reads are in — first place in LMArena's frontend code arena, beating Claude Fable 5 in 76% of direct duels, and fourth of 189 systems on the Artificial Analysis intelligence index, the strongest open-weights model measured. As of this writing the Hugging Face repository still shows an upcoming-release placeholder rather than downloadable files.

If

The weights land and self-hosting proves practical for enterprises, the pricing floor for frontier-class inference moves — and it moves from outside the American stack.

Why

Three things arrive together here and only one is a model. Testers report 70–80% of tokens consumed by internal reasoning, with complex runs taking up to an hour; Moonshot had to halt new subscriptions on a GPU bottleneck; and the White House has alleged both distillation of American models and circumvention of Nvidia export controls, which Moonshot denies. Frontier capability, brutal inference economics, and an export-control fight in one release.

Then

Watch the hardware floor rather than the leaderboard. A model that needs four to eight H100-class GPUs to run experimentally is open in licence and closed in practice — which is the whole inference-economics argument, stated by a Chinese lab this time. Our own Token Split Pulse already measures where the routed demand goes: in the latest daily reading, dated 25 July, Chinese models took 68.1% of tokens routed through the largest neutral model router against 30.7% for American models and 0.8% for European ones.

Token Split Pulse — Chinese vs US model usage →

What

Durable goods orders for June land at 12:30 UTC with consensus at +1.6% month-on-month against a prior of −4.5%; the ex-transportation cut is seen at +0.9% against +1.3% prior. The Dallas Fed manufacturing index follows at 14:30 UTC, consensus −1 against a prior of zero. After the close, Cadence Design is scheduled to report with consensus near $2.05 a share on roughly $1.58 billion, and Amkor Technology near $0.47 on roughly $1.81 billion.

If

The ex-transportation cut misses 0.9% — the core capital-goods proxy is the number that speaks to business investment, and it is the one that would confirm rather than contradict Friday's semi break.

Why

Cadence sits at the design layer, upstream of every chip cycle and inside our own EDA cohort, which fell 3.03% on Friday and 19.7% over the month. Amkor sits in packaging, the layer this letter has been watching for a warrant filing that would move it up a tier. Two prints, two ends of the same build-out.

Then

Nothing here is known until it prints. The consensus figures above are estimates and are stated as such — the read comes tonight, and it lands into a tape that has spent one session deciding the capital cycle is worth less.

What

The Federal Reserve decides Wednesday at 18:00 UTC with consensus for no change at 3.75%, followed by Chair Warsh's press conference at 18:30. Microsoft and Meta report the same day; Amazon and Apple on Thursday, alongside Q2 GDP at a consensus 2.3% and core PCE at +0.1% month-on-month, with the headline year-on-year seen easing to 3.7% from 4.1%. The two-year closed Friday at 4.33%, down from 4.37% on Thursday; the ten-year at 4.69%. The seven-to-ten-year Treasury fund reclaimed 93 on Friday at 93.03 after breaking it the session before.

If

Oil stays where it opened this morning, the September hike bet that peaked on Thursday has to come out of the curve — and the multiple compression it caused comes out with it.

Why

This is a hiking conversation, not a cutting one, and it is oil-driven. Singapore's central bank delivered a surprise back-to-back tightening on Sunday citing exactly that channel. A six-percent break in Brent two days before the decision changes the inflation arithmetic the committee is arguing about.

Then

Four capex guides in 48 hours are what Sunday's Weekly will be scored against. Alphabet lost $288.9 billion of market value after its print. The question is not whether the four beat — it is whether raised spending is punished the way Alphabet's and Intel's were.

What

Speaking to Bloomberg Intelligence over the weekend, Dan Ives framed the current tape as the third inning of the AI revolution, called Intel the comeback story of the cycle, and argued that for the first time in thirty years the United States is ahead of China in technology.

If

That last claim is taken as a testable proposition rather than a slogan, this morning offers three immediate tests of it — a Chinese DRAM champion up roughly 470% on debut with $85.5 billion of public capital behind its wafer plans, a Chinese lab releasing the strongest open-weights model ever measured, and our own token measurement showing Chinese models taking better than two of every three tokens routed on the open market.

Why

The friction is the point. The third-inning framing and the Friday tape cannot both be casually right: a market genuinely early in a capital cycle does not sell its best beat by eighteen percent. Either the tape is wrong about the innings, or the innings are being played by more teams than the framing assumes — and the leadership claim is the part of it we can actually measure.

Then

Read Ives in his own words, hold the two China headlines next to the leadership claim, and reach your own view.

Dan Ives · Yorkville Ives & Co ↗